Reproduce “PARSER: Read in Parallel, Reason in Depth for Long-Context LLM Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kun Li, Zexuan Qiu, Tianhua Zhang, Irwin King, Helen Meng
PARSER answers long-document multi-hop questions by sending focused queries to readers assigned to separate chunks in parallel, then using a lead agent to gather findings and ask follow-up questions. On HotpotQA, Qwen3.5-4B PARSER averages 84.57% Sub_EM across eight context lengths, versus 78.91% for ReMemR1. In these evaluations, its advantage is not uniform across tasks and lengths.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.